Machine-learning solutions present a privacy risk primarily because the training data used during the training phase may contain sensitive information. If this data is compromised, it can lead to privacy breaches. Machine-learning models can also inadvertently memorize and reproduce sensitive data from the training set.